def get_config_rnn(vocab_size, n=16):
config = net.Config(vocab_size)
config.embedding_dim = n
config.structure_type = 'rnn'
config.rnn_hidden_layers = 1
config.rnn_hidden_size = n
config.rnn_predict = True
config.rnn_type = 'blstm'
return config
def get_config_rnn(vocab_size):
config = net.Config(vocab_size)
config.embedding_dim = 200
config.structure_type = 'rnn'
config.rnn_type = 'blstm'
config.rnn_hidden_size = 200
config.rnn_hidden_layers = 2
# config.rnn_predict = True
return config
def get_config_rnn_large(vocab_size):
config = net.Config(vocab_size)
config.embedding_dim = 512
config.structure_type = 'rnn'
config.rnn_type = 'blstm'
config.rnn_hidden_size = 512
config.rnn_hidden_layers = 2
config.rnn_predict = True
config.rnn_share_emb = True
return config
def get_config_rnn(vocab_size, n=200):
from trf.common import net
config = net.Config(vocab_size)
config.embedding_dim = n
config.structure_type = 'rnn'
config.rnn_type = 'blstm'
config.rnn_hidden_size = n
config.rnn_hidden_layers = 1
config.rnn_predict = True
config.rnn_share_emb = True
return config
def get_config_cnn(vocab_size, n=32):
config = net.Config(vocab_size)
config.embedding_dim = n
config.structure_type = 'cnn'
config.cnn_filters = [(i, n) for i in range(1, 6)]
config.cnn_hidden = n
config.cnn_width = 3
config.cnn_layers = 3
config.cnn_skip_connection = True
config.cnn_residual = False
config.cnn_activation = 'relu'
config.cnn_batch_normalize = False
return config
def get_config_mix(vocab_size):
config = net.Config(vocab_size)
config.embedding_dim = 200
config.structure_type = 'mix'
config.cnn_filters = [(i, 128) for i in range(1, 11)]
config.cnn_hidden = 128
config.cnn_width = 3
config.cnn_layers = 3
config.cnn_activation = 'relu'
config.cnn_skip_connection = True
config.cnn_residual = False
config.cnn_batch_normalize = False
config.rnn_type = 'blstm'
config.rnn_hidden_size = 128
config.rnn_hidden_layers = 1
# config.rnn_predict = True
# config.rnn_share_emb = True
config.attention = True
return config
